Neighborhood Walkability and Body Mass Index Trajectories: Longitudinal Study of Canadians

Abstract

Objectives. To assess the impact of neighborhood walkability on body mass index (BMI) trajectories of urban Canadians.

Methods. Data are from Canada’s National Population Health Survey (n = 2935; biannual assessments 1994–2006). We measured walkability with the Walk Score. We modeled body mass index (BMI, defined as weight in kilograms divided by the square of height in meters [kg/m²]) trajectories as a function of Walk Score and sociodemographic and behavioral covariates with growth curve models and fixed-effects regression models.

Results. In men, BMI increased annually by an average of 0.13 kg/m² (95% confidence interval [CI] = 0.11, 0.14) over the 12 years of follow-up. Moving to a high-walkable neighborhood (2 or more Walk Score quartiles higher) decreased BMI trajectories for men by approximately 1 kg/m² (95% CI = −1.16, −0.17). Moving to a low-walkable neighborhood increased BMI for men by approximately 0.45 kg/m² (95% CI = 0.01, 0.89). There was no detectable influence of neighborhood walkability on body weight for women.

Conclusions. Our study of a large sample of urban Canadians followed for 12 years confirms that neighborhood walkability influences BMI trajectories for men, and may be influential in curtailing male age-related weight gain.

Over the past 3 decades, there has been a decline in physical activity and a rise in obesity prevalence worldwide.^1–3 The burdens of physical inactivity and obesity are recognized as major public health concerns because of their associated health risks,^4,5 accounting for substantial disability, health care utilization, and expenditure.^6,7 In light of the public health threats associated with the decline in physical activity and increase in body weight, there has been a growing interest and significant expansion of theoretical and empirical work investigating the underlying social and environmental causes of overweight and obesity.

One specific area of interest has been on the role of the built environment, including neighborhood walkability, in shaping routine, purposeful walking (often referred to as utilitarian walking) that, in turn, may favorably influence body weight. Utilitarian walking has shown significant associations with neighborhood walkability in a number of cross-sectional^8–16 and 2 longitudinal studies.^17,18 Previous cross-sectional studies have signaled geographical variations in body mass index (BMI, defined as weight in kilograms divided by the square of height in meters [kg/m²]) for men and women.^19–24 However, associations of BMI with different built environment measures have shown mixed results with the exception of urban sprawl, which has been consistently positively associated with BMI, and land use mix, which has been consistently negatively associated with BMI.^21,25

In Canada, there have been 2 cross-sectional studies examining associations between body weight and neighborhood walkability.^22,24 A study in Ontario used the Walk Score, which is a measure of local accessibility to amenities, and found that, on average, individuals who resided in the most walkable neighborhoods (Walk Score 90–100) weighed 3.0 kilograms less than those who lived in very car-dependent areas (Walk Score 0–24).²⁴ Negative associations between body weight and mixed land use, residential density, and average value of dwellings were also demonstrated in another study in Vancouver and Toronto metropolitan areas.²²

Few longitudinal studies have considered the relationship between a neighborhood’s walking-friendliness and BMI. One longitudinal study of older adults in the United States¹⁸ showed weak association: moving to a more walkable neighborhood (a 10-point-higher Walk Score) was associated with a 0.06 lower BMI (95% confidence interval [CI] = 0.12, 0.01). One study considered an adult population, but this was a local study in 1 metropolitan area (Edmonton, Alberta) with a modest follow-up time (6 years). This study found no significant association of BMI with neighborhood walkability.²⁶ Two studies in the United States followed a sample of adolescents who changed their residential neighborhood; the studies showed no association between living in a sprawling urban environment and BMI of youths,^27,28 although these 2 studies did not focus on neighborhood walkability per se.

Our study is the first to model adult BMI trajectories from a large population-based sample of adults in which the exposure of interest is neighborhood walkability. Our sample includes both movers (people who changed their residential neighborhoods) and nonmovers, giving us the opportunity to model change in BMI in relation to changes in neighborhood walkability by using residential relocations.

METHODS

Our sample comes from the National Population Health Survey (NPHS), a longitudinal survey conducted biannually by Statistics Canada starting in 1994–1995. The target population of the NPHS was household residents in the 10 Canadian provinces excluding some special groups (e.g., persons living on Indian Reserves and Crown Lands).²⁹ Access to the data was granted by the Social Sciences and Humanities Research Council of Canada (09-SSH-MCG-2068). Analyses were performed at the McGill–Concordia Quebec Inter-University Center for Social Statistics. We used the first 7 cycles of data collection, including baseline (1994–1995 to 2006–2007).

We restricted our analysis to young and middle-aged adults (aged 18–55 years at baseline) living in urban areas (≥ 50 000 population), who reported their weight and height. We included participants who either did not change their residential location or were relocated to a new neighborhood once during the follow-up period to allow for sufficient exposure time to neighborhoods with different walkability levels. We did not include individuals who moved more than once during follow-up to allow for sufficient exposure to the neighborhood environment. We excluded participants weighing less than 35 kilograms from the analysis.

Measures

The outcome measure was BMI. The NPHS respondents were asked to report their weight and height every cycle, and BMI was calculated by dividing their weight in kilograms by their height in meters squared (pregnant women were excluded). We modeled BMI as a continuous variable, which was normally distributed in our sample.

The primary exposure of interest was neighborhood walkability operationalized by the Walk Score (0–100). The Walk Score is based on distances to various weighted amenities (e.g., shopping, schools, parks, and restaurants). The measure has been validated against objective walkability measures^30,31 and has shown associations with BMI in a number of studies.^16,18 We divided the Walk Score into 4 quartiles. Low-walkable neighborhoods had scores from 0 to 39. Low-medium–walkable neighborhoods had scores from 40 to 55 and medium-high–walkable neighborhoods had scores from 56 to 69. High-walkable neighborhoods had scores from 70 to 100.

We constructed a variable that represented Walk Score quartiles for NPHS addresses (postal codes) at baseline, representing initial neighborhood walkability. We then constructed another variable by centering the Walk Score quartiles around their initial quartile level at baseline. This method is commonly used in longitudinal data analysis to measure change and allows for better interpretability of model estimates.³² From the centered Walk Score quartile variable, we constructed 4 time-varying dummy variables that indicated whether the respondents moved to a more walkable neighborhood (1 Walk Score quartile higher or ≥ 2 Walk Score quartiles higher) or moved to a less walkable neighborhood (1 Walk Score quartile lower or ≥ 2 Walk Score quartiles lower), from one survey cycle to another.

We stratified the BMI models by gender, following previous studies that have shown differences in associations of BMI with covariates for men and women.^20,28 We controlled for individual socioeconomic characteristics (age, education, marital status, and immigration status) and individual behaviors (leisure-time physical activity, utilitarian walking, and smoking status). Age was recorded at baseline as a continuous variable; education level was classified as having a postsecondary education (yes or no); marital status was classified as married, single, or divorced; and immigration status indicated whether the participants had immigrated to Canada in the past 5 years (at baseline; yes or no).

We calculated a physical activity index from leisure-time physical activity³³ and classified it as inactive (energy expenditure [EE] < 1.5 kcal/kg/day), moderately active (combined moderate EE = 1.5–2.9 kcal/kg/day) and active (EE ≥ 3 kcal/kg/day). Utilitarian walking measured the amount of walking per week to work, shopping, or other errands, but not for recreation (i.e., not leisure time), and we classified it into 4 categories (none, ≤ 1 hour, 1–5 hours, ≥ 6 hours). Smoking status had 3 categories: never smoker (≤ 25 lifetime cigarettes and current nonsmoker), former smoker (used to smoke daily or occasionally), and current smoker (daily or occasionally).

Statistical Analysis

We conducted an attrition analysis to ensure that participants lost to follow-up did not unduly influence sample characteristics.³⁴ We modeled BMI trajectories for men and women by using random-coefficient and fixed-effects regression models in Stata version 14 (StataCorp LP, College Station, TX). Random-coefficient regression models are also called growth curve models when time is the main covariate of interest.³²

We compared fixed-effects regression estimates with the random-coefficient regression estimates. Fixed-effects models eliminate bias resulting from unobserved heterogeneity caused by omitted confounders of time-constant covariates.³⁵ To account for the complex sampling design of the NPHS, we used population weights and bootstrap weights (for variance estimation), and compared these with the unweighted regression estimates.

RESULTS

The response rate to the NPHS was 92.8% in cycle 2, ending with 77.0% in cycle 7, with an average attrition rate of 2.3% across cycles. Our sample consisted of the NPHS respondents aged 18 to 55 years, living in urban areas, who moved once or did not move at all during the 12 years of survey follow up, and answered the survey at least twice (2 cycles). We dropped from the sample people who did not answer the survey starting from the second cycle. There were no significant differences in the mean health status, the mean leisure-time physical activity, or the mean BMI of people who were lost compared with those who remained in the sample (data are available upon request). The final sample was 2943 individuals with valid BMI for at least 2 cycles (1526 women and 1417 men). We mainly discuss findings for men as we did not find any influence of neighborhood walkability on BMI for women (Figure 1; Table A, available as a supplement to the online version of this article at http://www.ajph.org, reports results for women).

FIGURE 1—

Deriving the Analytical Sample: National Population Health Survey, Canada, 1994

Descriptive Statistics

The mean age of the 1417 men (48% of the overall sample) was 38 years (SD = 9). There were 371 men who changed their residential locations once and 1046 men who did not move. The proportion of men who were overweight or obese at baseline was 55%, and this increased to 61% at last follow-up. At baseline, the BMI of Canadian men was 1 kg/m² higher for individuals living in low-walkable neighborhoods (WSQ1) compared with their counterparts living in high-walkable neighborhoods (WSQ4; Table 1). The percentage of men with postsecondary education who moved from low- to high-walkable neighborhoods was 9% higher than those who moved from high- to low-walkable neighborhoods. The percentage of male immigrants was 15% lower among those living in low-walkable neighborhoods compared with those living in high-walkable neighborhoods. The percentage of married men was 28% higher among those living in low-walkable neighborhoods compared with those living in high-walkable neighborhoods. The percentage of men with children living in low-walkable neighborhoods was 30% higher than those living in high-walkable neighborhoods at baseline. More than half of men were inactive in their leisure time at baseline across all Walk Score quartiles.

TABLE 1—

Characteristics of Men at Baseline: National Population Health Survey, Canada, 1994

	Nonmovers (by Walk Score Quartile)				Movers Between Walk Score Quartiles
Variables	WSQ1a (n = 345), Mean (SD) or %	WSQ2b (n = 349), Mean (SD) or %	WSQ3c (n = 309), Mean (SD) or %	WSQ4d (n = 337), Mean (SD) or %	From Low to High (n = 91), Mean (SD) or %	From High to Low (n = 210), Mean (SD) or %
BMI, kg/m2	26.0 (3.9)	26.1 (3.8)	25.9 (4.1)	25.0 (3.8)	25.8 (3.5)	25.9 (4.0)
Age, y	39 (8.7)	38 (9.9)	38 (9.9)	37 (10)	37 (9.8)	36 (9)
Completed postsecondary education	59	55	54	58	62	53
Immigrants	16	21	27	31	28	17
Marital status
Married	78	71	66	50	66	65
Single	15	23	24	37	26	24
Divorced	7	6	10	12	8	11
Have children	74	67	61	44	64	54
Smoking status
Never smoker	36	37	38	39	40	40
Former smoker	29	30	29	27	24	30
Current smoker	34	32	33	35	36	30
Leisure-time physical activity
Inactive	55	57	62	53	60	56
Moderately active	26	24	21	26	20	29
Active	19	19	17	21	20	15
Utilitarian walking, per wk
None	50	45	47	38	46	35
Low (< 1 h)	14	17	18	17	20	20
Moderate (1–5 h)	18	22	19	32	15	30
High (≥ 6 h)	18	17	15	13	20	15

Note. BMI = body mass index; WSQ = Walk Score quartile.

^aLow-walkable neighborhood (scores 0–39).

^bLow-medium–walkable neighborhood (scores 40–55).

^cMedium-high–walkable neighborhood (scores 56–69).

^dHigh-walkable neighborhood (scores 70–100).

Around 50% of respondents who were living in low-walkable neighborhoods (WSQ1) did not report any utilitarian walking compared with 38% living high-walkable neighborhoods (WSQ4). The mean BMI for men (nonmovers) at survey follow-up intervals was patterned by neighborhood walkability; the lowest mean BMI at each time point was for those living in the high-walkable neighborhoods (WSQ4; Table 1).

Multivariate Analyses

Estimates from the weighted and unweighted regression models were similar for both the random-coefficient and fixed-effects regression models (Table B, available as a supplement to the online version of this article at http://www.ajph.org). In addition, there were minimal differences in the CIs of the weighted fixed-effects estimates, before and after we applied bootstrap weights. The bootstrap weights did not change the statistical significance of the variables suggesting that the complex design sampling nature of the NPHS did not induce any significant error in our sample. We discuss the unweighted random-coefficient estimates as they are more efficient (have smaller standard errors).³⁶

Interpreting the influence of time, age, and neighborhood walkability on body mass index.

Over each year of follow-up, BMI for men increased by 0.13 kg/m², regardless of baseline age. For every year increase in age at baseline, BMI increased by approximately 0.06 kg/m². At baseline, the mean BMI of men residing in neighborhoods with high walkability (WSQ4) did not have a statistically different BMI from those in living in low-walkability neighborhoods (WSQ1) but the point estimate was lower (−0.4 BMI; 95% CI = −0.95, 0.22). Moving from low- to high-walkable neighborhoods (2 Walk Score quartiles higher) was associated with approximately a 1-unit (kg/m²) decrease in BMI for men (95% CI = −1.7, −0.3). This effect is equivalent to 3 kilograms (approximately 6.8 lb) for a man of average height (178.2 cm). The estimates were consistent across weighted random-coefficient models (−1.10 BMI; 95% CI = −1.9, −0.36) and weighted fixed-effects models (−1.09; 95% CI = −1.77, −0.41). Moving from high- to low-walkable neighborhoods (2 Walk Score quartiles lower) was associated with an increase in BMI for men of approximately 0.45 kg/m² (95% CI = 0.01, 0.89; 0.4 kg/m² increase in BMI [95% CI = −0.11, 0.98] for the weighted random-coefficient estimate) compared with staying in a medium-high or high walkable neighborhood (WSQ3 and WSQ4). Weighted fixed-effects estimates were slightly higher—0.6 BMI increase (95% CI = −0.02, 1.22; Table B). We also tested the influence of neighborhood median household income, but it did not achieve significance, nor improve the model fit, and we dropped it from the models. The intraclass correlation showed that approximately 88% of the variance in the random parameters was explained by between-participant variance.

For women, over each year of follow-up, BMI increased by 0.13 kg/m², regardless of baseline age. For every year increase in age at baseline, BMI increased by approximately 0.09 kg/m². There was no statistically significant difference between baseline BMI of women by walkability quartile, nor with the change in walkability level as a result of moving (Table A).

Interpreting other covariates.

Moderate utilitarian walking (≥ 6 hours per week) was associated with 0.1 kg/m² lower BMI in men (95% CI = −0.21, 0.00), and approximately 0.17 BMI decrease (95% CI = −0.31, −0.03) after we eliminated bias resulting from time-constant omitted confounders. This is equivalent to an approximate 0.5 kilogram (1.1 lb) lower weight for a man of average height (178.2 cm). Current smokers had an estimated 0.43 lower BMI (95% CI = 0.29, −0.05) compared with never smokers. Those active in their leisure time had 0.17 lower BMI (95% CI = 0.02, 1.79) compared with less-active men. The BMI of single men was 0.45 lower (95% CI = −0.05, −0.34) than for married men. The BMI for recent immigrants (those who arrived to Canada 5 years or less before 1994 [cycle 1]) was approximately 1.1 kg/m² lower (95% CI = −1.60, −0.63) than for nonimmigrants. Completing postsecondary education was not associated with BMI.

Utilitarian walking did not have any significant association with BMI for women. Women who were active in their leisure time had 0.32 lower BMI (95% CI = 0.47, 0.18) compared with less-active women. The BMI of single women was 0.38 lower (95% CI = −0.70, −0.05) than for married women. The BMI for women who were recent immigrants was approximately 1.1 kg/m² lower (95% CI = −0.93, −0.20) than for nonimmigrants. Women who did not have children had a 0.57 lower BMI (95% CI = −1.63, −0.55) than women who had children (Table A).

Sensitivity Analysis

We estimated the predicted BMI of each survey respondent (men) at each point in time (i.e., based on individual and neighborhood characteristics) from the random-effects model (Table 2). We then calculated the average predicted BMI of different groups of men based on the walkability of their residential location.

TABLE 2—

Change in Body Mass Index Associated With Change in Neighborhood Walkability Estimated Through Random-Coefficient Models: National Population Health Survey, Canada, 1994–2006

Variables	Unweighted Coefficient Estimates (95% CI)
Body mass index, change kg/m2
Time	0.13 (0.11, 0.14)
Age centered around baseline mean age	0.06 (0.04, 0.09)
Baseline WSQ (Ref: baseline WSQ1)
Baseline WSQ2	0.31 (−0.24, 0.89)
Baseline WSQ3	0.23 (−0.36, 0.87)
Baseline WSQ4	−0.40 (−0.95, 0.22)
Change in WSQ (Ref: same WSQ)
Moved 1 WSQ higher	0.10 (−0.41, 0.51)
Moved 2 or 3 WSQs higher	–1.02 (−1.16, −0.17)
Moved 1 WSQ lower	0.19 (−2.07, −0.13)
Moved 2 or 3 WSQs lower	0.45 (0.01, 0.89)
Utilitarian walking (Ref: none)
Low	0.04 (−0.07, 0.15)
Moderate	−0.08 (−0.18, 0.02)
High	–0.11 (−0.21, 0.00)
Leisure-time physical activity (Ref: inactive)
Moderately active	−0.11 (−0.01, 0.84)
Active	–0.17 (−0.02, −1.79)
Smoking status (Ref: never smoker)
Former smoker	0.15 (−0.20, −0.01)
Current smoker	–0.42 (−0.29, −0.05)
Marital status (Ref: married)
Single	–0.45 (−0.05, −0.34)
Divorced	–0.21 (−0.70, −0.17)
Education level (Ref: completed postsecondary education)
Did not complete postsecondary education	−0.04 (−0.67, −0.22)
Recent immigrant (Ref: nonimmigrants)	–1.11 (−1.60, −0.63)
Constant	26.39 (25.95, 26.84)
Random-effects parameters
Standard deviation (AGEC+)	0.15 (0.14, 0.18)
Standard deviation (constant)	3.61 (3.37, 3.86)
Correlation (AGEC, constant)	0.15 (0.01, 0.28)
Standard deviation (residual)	1.38 (1.30, 1.47)
Intraclass correlation	0.87

Note. AGEC = age centered around the population mean; CI = confidence interval; WSQ = Walk Score quartile.

Figure 2 shows the trajectory curve of the predicted average BMI for the overall sample of men compared with average BMI trajectories for 4 groups of men: those who lived the entire follow-up period in low-walkable neighborhoods (WSQ1); those who lived the entire follow-up period in high-walkable neighborhoods (WSQ4); those who moved from low- to high-walkable neighborhoods (2 Walk Score quartiles change); and those who moved from high- to low-walkable neighborhoods (2 Walk Score quartiles change). The trajectory curves were presented as linear “curves” because there was no quadratic effect of time on BMI.

FIGURE 2—

Predicted Body Mass Index Trajectories for Men by Walk Score Quartile, Nonmovers and Movers: National Population Health Survey, Canada, 1994–2006

The average BMI (intercept) was 0.4 kg/m² (95% CI = −0.95, 0.22) lower for those who lived in high-walkable neighborhoods compared with those who lived in low-walkable neighborhoods. Moving to higher-walkable neighborhoods was associated with 1 kg/m² decrease in BMI (95% CI = −1.7, −0.3). Moving from high- to low-walkable neighborhoods was associated with an increase in BMI for men by approximately 0.45 kg/m² (95% CI = 0.01, 0.89) compared with those who did not move from their low-walkable neighborhoods.

DISCUSSION

Living in the most walkable urban Canadian neighborhoods (WSQ4) was associated with the lowest mean BMI for men. Moving from low- to high-walkable neighborhoods was associated with a reduction in BMI by approximately 1 kg/m², or 3 kilograms for a man of average height. Likewise, moving from high- to low-walkable neighborhoods was associated with an increase in BMI by approximately 0.45 kg/m². For women, BMI did not show any association with neighborhood walkability.

Our findings are consistent with Ross et al.²⁰ and Eid et al.²⁸ who acknowledged gender differences in the associations of BMI with the built environment characteristics. They also align with previous studies that found cross-sectional associations between neighborhood physical characteristics and BMI.^21,25 Hirsch et al.¹⁸ found weak associations between moving to more walkable neighborhoods (a 10-point-higher Walk Score) and BMI (reduction of 0.06 kg/m²; 95% CI = 0.12, 0.01). It is worth noting that Hirsch et al.¹⁸ were looking at older adults (aged 45–85 years at baseline), many of whom may not be interacting daily with their local built environment for utilitarian purposes.

Our findings contradict 2 previous longitudinal studies of neighborhood walkability and body weight. Eid et al.²⁸ did not find any influence of the built environment (measured as urban sprawl) on BMI of 4426 youths (aged 14–21 years) who responded to the US National Longitudinal Survey of Youth in 1979 and were followed for 7 years. The young age of the cohort, duration of follow-up (7 years), and the focus on urban sprawl (which is conceptually different from walkability), could be some of the underlying reasons why their findings did not align with ours. Berry et al.²⁶ followed 500 adults in Edmonton, Alberta, for 6 years and found nonsignificant associations of BMI with neighborhood walkability, although the change in BMI was associated with socioeconomic characteristics of neighborhoods.

We showed that moving to highly walkable neighborhoods (≥ 2 Walk Score quartiles higher) was associated with a reduction of BMI of 1 kg/m² (95% CI = −1.77, −0.41). The high-walkable neighborhoods had Walk Score values between 70 and 100 and represented neighborhoods similar to those in densely populated urban areas where one can access many amenities on foot. These types of neighborhoods were the ones that were associated with the lowest BMI trajectories for men. Interestingly, the second Walk Score quartile neighborhoods (Walk Score values between 40 and 55) were associated with the most unfavorable BMI trajectories, and not the least-walkable neighborhoods (those similar to the typical low-density suburban neighborhoods with Walk Score values between 0 and 39). One possibility is these least-walkable neighborhoods are actually reasonably well serviced by public transportation that induces utilitarian walking and lower BMI in their residents.¹⁵ Neighborhood walkability was an important predictor of male BMI trajectories even after we controlled for utilitarian walking. This suggests that there could be other factors such as neighborhood social norms that might influence body weight and are worth further exploration.

Our study relies on self-reported information about weight and height for the BMI calculation. Self-reported BMI is a clear limitation in cross-sectional studies. If, however, respondents overreport or underreport their weight or height, the direction of their misrepresentation is likely consistent over time, and hence reporting bias is arguably less of a problem in longitudinal analyses. Controlling for confounding variables to obtain precise coefficient estimates remains a problem even with longitudinal models. With fixed-effects regression (used in our analysis), bias in the error term that results from time-constant confounding variables is eliminated.^32,35,37

Another possible limitation of our work is the lack of availability of historical Walk Score data. We used 2012 Walk Score data, which did not correspond to the time frame of the NPHS follow up (1994–2006). That being said, neighborhoods do not usually change their physical characteristics quickly and we tested other measures correlated with Walk Score (street connectivity and population density) that we computed from street network and Census data in 1996 and 2006. Measures computed at the 2 time periods were highly correlated (Pearson correlation coefficient = 0.94; P < .01). To ensure that our results were not sensitive to our measure of walkability, we computed the same models with different walkability measures. The differences in mean predicted BMI with the 2012 Walk Score relative to other measures computed with 2006 street network, land use data, and Census data (e.g., population density [diff = 0.014; Pr( T > t ) = 0.4237], land use interaction measure³⁸ [diff = 0.014; Pr( T  > t ) = 0.4237], and walkability index³⁹ [diff = 0.0013; Pr( T > t ) = 0.9398]) were small and not statistically significant.

Lack of reliable nutrition variables in the NPHS is another limitation of this study. Nutrition information collection began in cycle 5 in the NPHS. We tested the amount of fruit and vegetable consumption as a potential predictor of BMI and it did not influence the effect of neighborhood walkability on BMI trajectories for men. We have chosen not to use the nutrition variable so as not to lose the full range of years to predict BMI trajectories.

We demonstrated a clear signal of the influence of moving to both higher- and lower-walkable neighborhoods on male BMI trajectories, even after we controlled for unobserved heterogeneity from time-constant omitted confounders. Our findings suggest that neighborhood walkability is an important factor in curbing the population-level rise of BMI with age for men, and that men who move to highly walkable places enjoy a BMI advantage over time. Given that there have been so few policy options for the obesity epidemic that have had widespread success, these results are compelling for considering built environment modifications among policy options for obesity control in populations.

Associations of neighborhood walkability with BMI persisted after we controlled for many individual-level covariates, including utilitarian walking. Our findings for men suggest that “normal” age-related weight gain can be moderated by living in highly walkable neighborhoods. Understanding the precise mix of neighborhood attributes (both physical and social) that are associated with reductions in body weight would be useful to direct the types of environmental modifications that could be implemented to link urban planning policy more directly with health policy, consistent with the World Health Organization’s 2010 Kobe statement “to integrate health and health equity in all urban public policies.”^40(p2)